Will foundries bite the release schedule of the XBox Next, or will Microsoft keep the dates they have told those in the know? The recent rash of delay rumors, backed up by some very unusual moves make this more than an idle question.
SemiAccurate has been hearing about problems at Microsoft with their next console part since we exclusively reported the tapeout. The short story is that officially the PoR was December 2012 for production wafers in (Note: This is a specific term with a very specific meaning, if you don’t understand what it implies, SemiAccurate does do technical consulting) has not changed. The rumors however are growing much more persistent and frequent, so we spent the last month digging in to the details.
The chip, which is still referred to as ‘Oban’, is being run through multiple fabs in very high quantities, too high by more than an order of magnitude to simply be for dev kits. Yields on the chip are said to be something between painfully low, Nvidia Fermi painfully low, and worse than that. Given the sheer number of wafers Microsoft contracted for, this seems to be both an anticipated problem, and one they have plans to work through. That said, SemiAccurate’s sources are still reporting that there is much work to be done, yields are not even up to “horrid” yet.
One of the things Microsoft seems to be trying is dual or triple sourcing the parts, hoping the elves at one foundry will fix things ahead of the rest. Even if none do, three chances of a breakthrough are better than one. While there is still time to get things right, it takes about 8 weeks to run a hot lot through a fab and analyze the results. If production wafers are really going in this year, the last time a foundry can run a test wafer is about November 1. Anything after that day will not come out of the fab in time for the results to be fed back in to the production chips. That is the technical term for “The XBox Next is going to be delayed”.
Sources on the supply chain side of the console world tell SemiAccurate that there is some slack in the production schedule, six to eight weeks of delay would be possible before the launch would have to be pushed back. That meshes well with what SemiAccurate knows about silicon production times and ODM capabilities. Microsoft insiders tell us that the planned launch date is September 2013, and that is not changing without heads rolling internally.
So, what is the XBox Next? SemiAccurate has been saying for a while that all signs were pointing toward a PowerPC, specifically an IBM Power-EN variant. The dark horse was an x86 CPU, but it was a long shot. It looks like the long shot came through, moles are now openly talking about AMD x86 CPU cores and more surprisingly, a newer than expected GPU. How new? HD7000 series, or at least a variant of the GCN cores, heavily tweaked by Microsoft for their specific needs.
This means both the XBox Next and the PS4 are going to effectively be HSA/FSA devices. Stop and think about that for a minute, AMD has the GPUs for all three next generation consoles, the CPUs for both the XBox Next and the PS4, and is effectively controlling the middleware/interpreted language that is the plumbing for all of it. No matter what the future of video gaming is, PC, console, or mobile, it will almost assuredly have AMD at its center. Yes, even Apple.
Getting back to the premise of the story, the rumored delays in the next XBox console, we have a lot of inputs. High level executives bailing out of AMD at a pace rapid enough to wear tracks in the carpet outside of the HR department, Microsoft running around testing every fab that they can get access to, and a shroud of secrecy surrounding the whole thing. SemiAccurate has heard everything from vendor negotiating tactics to last minute design changes for the purported delays, and has tracked them all down to knowledgeable sources.
In the end, there is one, and only one fact at the center of all of these rumors, semiconductor yields. The exact details are harder to come by, but given the very hard limits imposed by the fab technologies involved and the self-imposed start date, time is running out. If there is two months of slack in the supply chain, unless the yield problems are solved by late February 2013, Microsoft can’t hit it’s desired September 2013 launch date.
With that in mind, the last chance any foundry has to start a test wafer is ironically the same end of December date that Microsoft wanted volume production to start. Luckily for everyone involved, breakthroughs can happen at literally any time, so there is still lots of hope. Redmond is telling everyone downstream of them that there is no problem, no delay, and everything is well. While that doesn’t seem to be a realistic perspective from SemiAccurate’s point of view, it isn’t incorrect either. Yet. Lets hope Microsoft can work out the silicon bugs, the industry really needs a boost.S|A
Latest posts by Charlie Demerjian (see all)
- SemiAccurate has Skylake-SP die shots, sizes, and more - Jul 21, 2017
- Skylake-SP has a diverged core - Jul 19, 2017
- Intel’s Purley platform architecture is a step forward - Jul 13, 2017
- Intel launches Purley aka metal Xeons - Jul 11, 2017
- AMD’s Epyc has lots of connections - Jun 26, 2017